Financial Inclusion and Development
Der Wirtschaftswissenschaftlichen Fakult¨at der Gottfried Wilhelm Leibniz Universit¨at Hannover
zur Erlangung des akademischen Grades
Doktorin der Wirtschaftswissenschaften - Doctor rerum politicarum -
genehmigte Dissertation
von
M.A. Theres Kl¨uhs
geboren am 28.03.1989 in Zwickau
2019
Koreferentin: Apl. Prof. Dr. Susan Steiner Tag der Promotion: 08.07.2019
This dissertation contributes to enhancing the knowledge about financial inclusion by focusing on its consequences in emerging economies. Chapter 1 emphasizes the importance of financial inclusion, defined as access to and use of financial services, and relates the different chapters to the overall topic. Chapter 2 analyzes whether the mode of providing access to finance itself changes financial behavior and business outcomes. It analyzes a randomized controlled trial which provided a one-time cash transfer to micro and small entrepreneurs in Kampala, Uganda, in 2013. One half of the treatment group received the transfer in cash, the other half had the money transferred on their bank accounts, which we assume to work as a soft commitment device inducing entrepreneurs to rather use the money for business related expenses. We do not find any direct effect of the transfer on monthly profits and capital stock. However, we detect positive short-term treatment effects on the more “upstream” variables inventories and sales for entrepreneurs in the account treatment group.
Chapter 3 analyzes determinants of financial inclusion and focuses on financial lit- eracy as a possible demand side driver of inclusion. It combines cross country data on financial literacy with information on financial inclusion, financial infrastructure, and other country characteristics. We establish a robust positive relation between financial literacy and four different dimensions of financial inclusion. Considering institutional variation across countries and regarding “access to finance”, financial literacy and finan- cial infrastructure mainly substitute each other. With regards to the “use of finance”, higher financial literacy strengthens the effect of more financial depth. To respond to reverse causality concerns, we employ an instrumental variable strategy, which sup- ports a causal interpretation of our results. Further robustness checks do not alter the findings either.
Last, Chapter 4 points out a possible drawback of using financial services. It asks whether too high expectations regarding future income may actually harm households and lead them to accumulate too much debt. We collect extensive data on debt and borrowing behavior of households in rural Thailand which enable us to calculate both expectations about future monthly income as well as various debt indicators. Control- ling for specific household characteristics, we find a strong relationship between our two measures of biased expectations and (over-)indebtedness. The more quantitative expec- tation measure is stronger related to objective debt and the more subjective expectation measure is rather related to our subjective debt measure. An additional lab-in-the-field experiment shows that over-confidence is indeed related to over-borrowing.
i
ii
Writing this dissertation has taught me in various ways for a life-time and I owe thanks to many without whom this thesis would not exist.
First of all, I would like to thank Lukas Menkhoff, my first supervisor, for his support and guidance at all stages of my doctoral studies. Continuously, he took time to listen carefully to both good and challenging parts of my research while encouraging me to stay focused, to formulate precisely, and to work in a solution-oriented manner. Under his supervision, I could work under very fair and trusting conditions, which enabled me to reconcile work and family life in an extraordinarily good way. Thank you.
I also thank Susan Steiner, my second supervisor, for her advice and comments on my academic work. She combined apt and well-intentioned criticism with an ever- friendly and pleasant attitude. This kind of atmosphere made discussing even the flaws of my work with her an enjoyable experience. Furthermore, I thank Patrick Puhani for his readiness to be the third supervisor.
The thesis consists of three papers all of which are joint work with co-authors. I thank Antonia Grohmann, Melanie Koch, Lukas Menkhoff, Tevin Tafese, and Wiebke Stein for the opportunity to develop research ideas and output together. I learned a lot from collaborating with you! I also thank the German Research Foundation (Grant RTG 1723 and CRC 190), the Thailand Vietnam Socio Economic Panel (TVSEP), and the German Institute of Global and Area Studies (GIGA) for their generous funding and/or support which, among others, provided the basis to pursue field-work in Uganda and Thailand.
I am grateful to be part of the research training group “Globalization and Develop- ment” and would like to thank all my colleagues many of whom have become friends over the years. You made this journey a lot more enjoyable. Likewise, I thank my other colleagues at the Leibniz University Hanover for shared coffee times, inspiring conversations and the opportunities to build friendships.
The dissertation would have not been possible without the ongoing support of my family. Thank you, Stephan, for encouraging me always and for being the best husband and father one could wish for! I also wish to thank my parents, sister, and family in law - you have all supported this work in various ways.
Last, I am grateful to God beyond measure - it is only by His endless grace that I have started, continued and finished this work.
iii
Table of Contents vi
List of Figures viii
List of Tables xiii
1 Introduction 1
2 Rethinking the Effectiveness of Cash Transfers - Evidence from a
Field Experiment in Uganda 7
2.1 Introduction . . . 8
2.2 Background and Study Design . . . 13
2.3 Data . . . 16
2.3.1 Measurement of Main Variables . . . 16
2.3.2 Sample Attrition . . . 17
2.3.3 Summary Statistics and Balance Tests . . . 18
2.3.4 Self-Reported Evidence on Use of Cash Transfer . . . 20
2.4 Results . . . 21
2.4.1 Estimation Strategy . . . 21
2.4.2 Estimation of Basic Experimental Treatment Effects . . . 22
2.4.3 Heterogeneity of Treatment Impacts . . . 24
2.5 Robustness Tests and Discussion . . . 26 v
Evidence 51
4 Don’t Expect too Much: The Effect of Biased Expectations on (Over)-
Indebtedness 53
4.1 Introduction . . . 54
4.2 Data . . . 57
4.2.1 The Thailand Vietnam Socio Economic Panel . . . 57
4.2.2 Income Expectation Biases . . . 58
4.2.3 (Over-)Indebtedness Indicators . . . 63
4.2.4 Descriptive Statistics . . . 64
4.3 Survey Results . . . 66
4.3.1 Estimation Strategy . . . 67
4.3.2 Main Results . . . 67
4.3.3 Robustness . . . 71
4.4 The Experiment . . . 72
4.4.1 Experimental Design . . . 73
4.4.2 Experimental Results . . . 76
4.4.3 Confounding Factors . . . 79
4.5 Conclusion . . . 81
Bibliography 109
Appendix A - Appendix for Chapter 2 111
Appendices B and C - Appendices for Chapter 3 128
Appendix D - Appendix for Chapter 4 181
vi
2.1 Use of Transfer - Cash vs. Account . . . 49
2.2 Use of Transfer - Men vs. Women . . . 49
4.1 Household Debt to GDP Ratio, Selected Emerging Markets . . . 83
4.2 Study Site, Ubon Ratchathani Thailand . . . 83
4.3 Sampled Subdistricts . . . 83
4.4 Probability Density Function of Expected Income . . . 84
4.5 Number of Loans . . . 84
4.6 Income Certainty . . . 85
4.7 Experimental Flow . . . 95
4.8 Cumulative Density Distribution of Expected Rank by Treatment . . . 96
4.9 CDFs of Self-Confidence . . . 96
4.10 Histogram for Self-Confidence . . . 96
4.11 Mean Expected Rank by Treatment . . . 96
4.12 Mean Consumption by Treatment . . . 96
C.1 Average Marginal Effects of Financial Literacy on Four Measures of Financial Inclusion at Different Levels of GDP per capita. . . 176
C.2 Average Marginal Effects of Financial Literacy on Four Measures of Financial Inclusion at Different Levels of Bank Branches per 1000 km2. 177 C.3 Average Marginal Effects of Financial Literacy on Four Measures of Financial Inclusion at Different Levels of Private Credit to GDP (IV). . 178
vii
(IV). . . 179 C.5 Average Marginal Effects of Financial Literacy on Four Measures of
Financial Inclusion at Different Levels of GDP per capita (IV). . . 180 D.1 CDF for the Expected Rank by Treatment, After the Main Quiz . . . . 185
viii
2.1 Industry Sampling Categories . . . 30
2.2 Assigned Treatment for Baseline and Follow-up Waves . . . 30
2.3 Analysis of Attrition - Intended Treatment . . . 31
2.4 Baseline Balance - Account vs. Control Group . . . 32
2.5 Baseline Balance - Cash vs. Control Group . . . 33
2.6 Baseline Balance - Account vs. Cash Group . . . 34
2.7 Use of Cash Transfer - Account vs. Cash Group . . . 35
2.8 Use of Cash Transfer - Men vs. Women . . . 35
2.9 Impact of Cash Transfer on Capital Stock . . . 36
2.10 Impact of Cash Transfer on Monthly Profits . . . 36
2.11 Impact of Cash Transfer on Inventories . . . 37
2.12 Impact of Cash Transfer on Total Sales . . . 37
2.13 Impact of Cash Transfer on Total Savings . . . 38
2.14 Impact of Cash Transfer on Total Business Savings . . . 38
2.15 Het. Treatment Effects w.r.t. Bus. Savings - Capital Stock . . . 39
2.16 Het. Treatment Effects w.r.t. Bus. Savings - Profits . . . 40
2.17 Het. Treatment Effects w.r.t. Bus. Savings - Inventories . . . 41
2.18 Het. Treatment Effects w.r.t. Bus. Savings - Sales . . . 42
2.19 Het. Treatment Effects w.r.t. Any Savings - Capital Stock . . . 43
2.20 Het. Treatment Effects w.r.t. Any Savings - Profits . . . 44 ix
2.23 Impact of Cash Transfer on Monthly Profits - Lee Bounds . . . 46
2.24 Impact of Cash Transfer on Capital Stock - Lee Bounds . . . 47
2.25 Impact of Cash Transfer on Total Savings - Lee Bounds . . . 48
4.1 Probabilities Assigned to Sections of the Income Distribution . . . 84
4.2 Summary Statistics - Main Variables . . . 85
4.3 Correlation Matrix - Debt Variables . . . 86
4.4 Income Expectation Bias Dummy - Objective Debt Indicators . . . 87
4.5 Income Expectation Bias Dummy - Subjective Debt Indicators . . . 88
4.6 Income Expectation Bias Dummy - Over-Indebtedness Indicators . . . 89
4.7 Fin. Forecast Error - Objective Debt Indicators . . . 90
4.8 Fin. Forecast Error - Subjective Debt Indicators . . . 91
4.9 Fin. Forecast Error - Over-Indebtedness Indicators . . . 92
4.10 Certainty Measure - Objective Debt Indicators . . . 93
4.11 Certainty Measure - Subjective Debt Indicators . . . 93
4.12 Certainty Measure - Over-Indebtedness Indicators . . . 94
4.13 Descriptive Statistics across Treatments . . . 95
4.14 Consumption Decision . . . 97
4.15 Overborrowing and Overspending . . . 97
4.16 Overborrowing in the Game and in Real Life . . . 97
A.1 Actual Treatment for Baseline and Follow-up Waves . . . 111
A.2 Analysis of Attrition - Baseline Balance . . . 112
A.3 Het. Treatment Effects w.r.t. Gender - Capital Stock . . . 114
A.4 Het. Treatment Effects w.r.t. Gender - Profits . . . 115
A.5 Het. Treatment Effects w.r.t. Gender - Inventories . . . 116 x
A.7 Het. Treatment Effects w.r.t. Gender - Any Savings . . . 118
A.8 Het. Treatment Effects w.r.t. Gender - Business Savings . . . 119
A.9 Het. Treatment Effects w.r.t. High Education - Capital Stock . . . 120
A.10 Het. Treatment Effects w.r.t. High Education - Profits . . . 121
A.11 Het. Treatment Effects w.r.t. High Education - Inventories . . . 122
A.12 Het. Treatment Effects w.r.t. High Education - Sales . . . 123
A.13 Het. Treatment Effects w.r.t. High Education - Any Savings . . . 124
A.14 Het. Treatment Effects w.r.t. High Education - Business Savings . . . . 125
B.1 Financial Literacy Questions and Response Options . . . 128
B.2 Control Variables Summary Statistics and Sources . . . 130
B.3 Correlations between Control Variables . . . 132
B.4 List of Countries in OLS and IV Regressions . . . 133
B.5 Correlations between Financial Literacy and Outcome Variables . . . . 134
B.6 Basis for Imputations for Numeracy in Primary School . . . 135
B.7 First Stage Regressions for IV Results . . . 136
B.8 Financial Literacy and Access to Financial Services - OLS and IV Results137 B.9 Financial Literacy and Use of Financial Services - OLS and IV Results 138 B.10 First Stage Regression for Placebo IV Results – Literacy as an Instrument139 B.11 First Stage Regression for Placebo IV Results – Literacy as an Instrument140 B.12 Financial Literacy and Access to Finance - OLS Results, Proportion of Banks that are State Owned . . . 141
B.13 Financial Literacy and Access to Finance - OLS Results, Proportion of State Owned Assets Below the Median Only . . . 142
B.14 Financial Literacy and Access to Finance - IV Results, Controlling for Government Expenditure and Total Education Expenditure . . . 143
xi
B.16 Financial Literacy and Access to Finance - IV Results, Using 1960s Numeracy . . . 145 C.1 Financial literacy and Access to Finance: IV Results using Lewbel (2012)155 C.2 Financial literacy and Use of Financial Services: IV Results using Lewbel
(2012) . . . 156 C.3 Financial Literacy, GDP, and Their Interaction . . . 157 C.4 Financial Literacy, Bank Branch Penetration, and Their Interaction . . 158 C.5 Financial Literacy and Financial Inclusion for Different Income Groups 159 C.6 Financial Literacy and Financial Inclusion - OLS, Excluding Countries
with More than 50% Muslims . . . 160 C.7 Financial literacy and Financial Inclusion - IV, Excluding Countries with
More than 50% Muslims . . . 161 C.8 Financial Literacy and Borrowing Decisions - OLS and IV Results . . . 162 C.9 Financial Literacy and High Frequency of Use - OLS and IV Results . . 163 C.10 Financial Literacy and Account Ownership - Additional Control Variables164 C.11 Financial Literacy and Financial Services - OLS Results Including Coun-
try Group Dummies . . . 165 C.12 Financial Literacy and Access to Financial Services - OLS Results -
Without Population And/Or Education Variables . . . 166 C.13 Financial Literacy and Use of Financial Services - OLS Results - Without
Population And/Or Education Variables . . . 167 C.14 Fin. Literacy and Fin. Inclusion Incl. Religiosity - OLS Results . . . 168 C.15 Financial Literacy and Financial Inclusion Incl. Hofstede’s Cultural Di-
mensions – OLS results . . . 169 C.16 Financial Literacy and Financial Inclusion Incl. Legal Origin – OLS Results170
xii
C.18 Financial Literacy and Financial Inclusion - Quantile Regressions . . . 172
C.19 Financial Literacy, Financial Depth and Their Interaction (IV) . . . 173
C.20 Financial Literacy, Bank Branch Penetration and Their Interaction (IV) 174 C.21 Financial Literacy, GDP, and Their Interaction (IV) . . . 175
D.1 Subsample: Income Expectation Bias Dummy - Objective Debt Indicators181 D.2 Subsample: Income Expectation Bias Dummy - Subjective Debt Indicators181 D.3 Subsample: Income Expectation Bias Dummy - Over-Indebt. Indicators 182 D.4 Wider and Narrower Bias Measures - Objective Debt Indicators . . . . 182
D.5 Wider and Narrower Bias Measures - Subjective Debt Indicators . . . . 182
D.6 Wider and Narrower Bias Measures - Over-Indebtedness Indicators . . 183
D.7 All Biases - Incl. Lagged Dependent Variable . . . 183
D.8 All Biases - Interaction with Conscientiousness . . . 184
D.9 Descriptive Statistics by Participation in Game . . . 185
D.10 Linear Probability Model Participation in Game . . . 186
D.11 Descriptive Statistics for Excluded Sample . . . 186
D.12 Descriptive Statistics for Non-Rationals (only significant effects reported)187 D.13 Descriptive Statistics for Rationals (only significant effects reported) . . 188
D.14 Consumption Decision including Rationals . . . 188
xiii
Introduction
The life of money-making is one undertaken under compulsion, and wealth is evidently not the good we are seeking; for it is merely useful and for the sake of something else.
- Aristotle, The Nichomachean Ethics, I.5 -
Money is one of the key prerequisites for leading a good life. Individuals who face a lack of money do not just fall short of sufficient finances, but they are restricted in their freedom to attain the desired set of functionings that they deem important for their lives (Sen, 1999, 2003). Clearly, then, finances are ‘not the good we are seeking’ but they are ‘for the sake of something else’. In the same manner, the proficient management of finances is not an end in itself. Rather, the ability to save, invest, borrow, and to take out insurances is an important functioning that helps individuals increase their freedom and live the lives they envision.
Most individuals in industrialized countries are able to conveniently deal with fi- nancial matters using a diverse range of services offered by banks, insurances, and other financial service providers. However, access to such services is far from universal in developing and emerging economies. Households in these economies have developed sophisticated coping mechanisms in response to the lack of formal financial services (Collins et al., 2009; Banerjee and Duflo, 2012). Yet, these mechanisms cannot fully offset the effect of not having access to formal financial infrastructure. Therefore, indi- viduals face greater risks with regards to unexpected shocks, they are constrained in taking out loans keeping them from investing in their education, health, or businesses, and their ability to save is limited due to insufficient safe storage places for money
1
On a global scale, access to formal financial services is propagated as a key enabler to achieving various of the 17 Sustainable Development Goals. Eradicating poverty (SDG 1), ending hunger (SDG 2), achieving good health (SDG 3) and gender equality (SDG 5) might be the most prominent goals that are connected to having access to finance (Klapper et al., 2016).
Financial inclusion, defined as access to and use of formal financial services, has hence been pushed forward by both researchers and high level policy makers alike such as the G20 via the Global Partnership for Financial Inclusion, the United Nations, the World Bank, and the Alliance for Financial Inclusion. And indeed, there is good news.
Since 2011, the number of adults owning a bank account worldwide has increased from 51 percent to 69 percent in 2017 (Demirg¨u¸c-Kunt et al., 2018). These are 1.2 billion adults who have gained access to financial services since 2011. The unprecedented surge in digital financial access contributes a fair share to this development. For example, studies related to digitized payment systems find more efficient service delivery and higher trust in the transfer providers and granting access to formal digital savings products indicates positive household welfare impacts (Demirg¨u¸c-Kunt et al., 2017;
Karlan et al., 2016).
Nevertheless, about 1.7 billion people remain unbanked (Demirg¨u¸c-Kunt et al., 2018). The World Bank Group has, therefore, committed to the “Universal Financial Access Initiative” ambitiously envisioning that all adults worldwide will be able to access a formal bank account by the year 2020.
Simultaneously, a new frontier with respect to financial inclusion is emerging: It is to move beyond access, to reach an active and informed usage of financial products.
Currently already, one fifth of formal bank accounts are dormant and two-thirds of mobile money accounts are not used regularly (Demirg¨u¸c-Kunt et al., 2018).
All initiatives for enhanced access to finance will fail to make a long-lasting impact, unless proponents of financial inclusion ensure that customers are able to make an informed and beneficial use of products, which are tailored to their needs and well regulated. Therefore, improved access to available financial products must be conjoined with capacity building on how to use them in order to help customers make informed decisions.
This dissertation contributes to advancing the understanding of financial inclusion with special focus on its consequences in emerging economies. First, it considers the effects of access to finance on business outcomes of micro and small enterprises (MSEs) (Chapter 2). It analyzes whether a financial product itself changes financial behavior
and business outcomes, thereby adding a new finding to the literature on cash transfers in the business realm. The subsequent chapter elaborates on the individual skill set that is beneficial for successful financial inclusion. On a macro level, it shows that financial literacy is an important driver of both access toand use of finance (Chapter 3). The last chapter focuses solely on theusage side and points out downsides of financial inclusion processes (Chapter 4). Due to their focus on the demand side, chapters 3 and 4 add an important and understudied aspect to the discussion of financial inclusion.
Road Map
Chapter 2, co-authored with Tevin Tafese, examines the effect of accessing formal finance for MSEs in Uganda. Providing access to finance for entrepreneurs follows the rationale that businesses have great growth potential because they likely realize high marginal returns to capital in response to the injection of financial means into the enterprises. Cash transfers are a way to alleviate possible constraints and have been thoroughly studied (see, for example, with respect to MSEs de Mel et al., 2008, 2012;
Fafchamps et al., 2014; Berge et al., 2015; Blattman et al., 2016).
We add to the existing body of research by conducting a randomized controlled trial, which has a special twist. In addition to paying transfers, we vary the mode of receiving the money and analyze whether enterprises prosper relatively more upon obtaining a transfer into their bank account contrary to receiving the money in cash.
Thereby, we implicitly assume the bank account to work as a commitment device, i.e.
beneficiaries are more likely to use the money for business related expenses as opposed to on-the-go purchases, which might rather be made when the money is received in cash.
Hence, Chapter 2 ultimately evolves around the question if access to formal financial services and products themselves may drive positive changes in financial behavior.
While we do not find any overall direct effects of the transfer on profit and capital stock, we detect positive treatment effects in the short term on more “upstream” business variables, specifically on inventories and sales. For these outcomes, the bank account indeed serves as a commitment device altering financial behavior because the effects are only traceable for the ‘account treatment’ group and not for those who received the money in cash.
Next, the dissertation turns to analyzing determinants of financial inclusion. Re- search at the country level has shown that better financial inclusion is related to char- acteristics such as more financial depth, clearer legal requirements, or low cost banking services (Allen et al., 2016). These determinants are solely related to the supply side of financial markets. Yet, functioning financial markets need not only adequate infras-
customers will more likely make sophisticated financial decisions and demand higher inclusion themselves. Therefore, Chapter 3 studies financial literacy and its role in advancing financial inclusion at the country level. This becomes feasible thanks to combining data on financial inclusion from the Global Findex Database with newly available macro data that contain information about the state of financial literacy in 143 countries (Klapper et al., 2015; Demirg¨u¸c-Kunt et al., 2018).
We contribute twofold to the literature. First, we indeed find a positive relation be- tween financial literacy and four measures of financial inclusion. Two of these measures are rather concerned with providing access to financial services (account ownership, debit card ownership). The other two measure the extent to which these services and products are used (actively saving at a formal financial institution, using a debit card in the last year). The relation persists even when controlling for institutional and country characteristics. Second, financial literacy has a differing effect depending on the type of financial inclusion. Regarding the access to finance, the marginal effect of financial literacy decreases with higher financial depth, i.e. literacy can be substituted with a better institutional environment. However, with regard to the use of financial products, financial literacy and depth complement and even re-inforce each other. To respond to reverse causality concerns, we employ an IV strategy, which supports a causal interpre- tation of our results. In conclusion, we show that an informed client base equipped with sufficient financial literacy is as important to advancing the cause of financial inclusion as is increasing financial depth. The chapter is joint work with Antonia Grohmann and Lukas Menkhoff and has been published in World Development.
While Chapter 3 finds that the individual skill set matters to benefit from access to and use of finance, the last chapter points out a possible drawback of supply side led financial inclusion. Analyzing data from rural households in Thailand, it examines how overly positive expectations regarding future household income drive current debt levels and the likelihood to be over-indebted. Chapter 4 is joint work with Melanie Koch and Wiebke Stein.
Thailand is a prime example to study the consequences of financial access, as more than 80% of the population have quite recently gained access to bank accounts (Demirg¨u¸c-Kunt et al., 2018). We elicit extensive data that allows us to compile differ- ent debt indicators and to precisely calculate subjective expected household income, which includes measures on different moments of the expected income distribution.
We find a robust relationship between biased income expectations and (over)- indebtedness. The more income households expect to generate in the future, the more
debt they accumulate, even to a point where it becomes detrimental to their financial health. While accumulating debt with rising income is common and desirable, expec- tations about rising income do not account for the possibility that the future income will fail to rise as high as expected - making households more likely to become over- indebted.
The results vary with respect to different debt and expectations indicators. ‘Hard’
or objective debt measures, such as the debt service to income ratio, are more affected by the subjective expectation bias, which compares the amount of current and future expected income. In contrast, the qualitatively elicited financial forecast error is more strongly associated with ‘soft’ or subjective debt indicators measuring financial distress rather than actual debt. These results persist also with various robustness specifica- tions. Additionally, certainty about future income affects debt holdings, too. The more certain respondents are about their income expectations, the more debt they accu- mulate. Paired with the uncertain environment the households are situated in, this increases the likelihood of falling into debt.
A lab-in-the-field experiment underpins these results by showing that overconfi- dence regarding income is systematically related to over-spending. However, it cannot establish causality because the exogenously varied level of self-confidence does not result in a significant difference regarding the propensity to over-borrow. The paper contributes to the small but growing literature on (over-)indebtedness and its drivers.
From a policy perspective, the findings call for building and improving knowledge about financial products on the one hand, and about households’ financial situations on the other hand.
In conclusion, this dissertation offers insights into financial inclusion in emerging economies. There remain two contributions particularly noteworthy to the reader and for the advancement of access and use of financial services in general. First, not only the possibility toaccess finances bears positive outcomes on development but the mode of accessing these resources alters financial decisions themselves. This may have far reach- ing implications regarding the design of financial products. For example, government transfers could be rolled out on bank accounts such that households’ savings behav- ior is influenced positively. Moreover, formal financial service providers should develop products conducive to their clients’ ‘financial health’ - especially in the context of emerging economies, where access to finance has spread only recently (Demirg¨u¸c-Kunt et al., 2017). Second, the skills and financial behavior of customers matter for the use of financial infrastructure. Hence, from a demand side point of view, policies should center around fostering financial knowledge and transparently communicating benefits
built and a deepening of financial inclusion beyond access becomes feasible.
Rethinking the Effectiveness of Cash Transfers - Evidence from a Field Experiment in Uganda 1
with:
Tevin Tafese
1 This paper builds upon the author’s master thesis. We would like to thank seminar participants in Hannover and Hamburg for helpful comments and suggestions on this essay, in particular Lena Giesbert, Stephan Klasen, Jann Lay, Lukas Menkhoff, Sebastian Prediger, and Helke Seitz.
7
Cash transfer programs have spread rapidly across over 120 countries in recent years and are a popular tool to fight poverty (World Bank, 2018a). The expansion of cash transfer programs has been accompanied by a growing body of studies evaluating their efficacy.2 Indeed, most of them positively affect a broad range of outcomes, such as children’s educational attainment (Baird et al., 2014), access to and use of health care facilities (Cahyadi et al., 2018), nutritional status (Manley et al., 2013), and household consumption (Haushofer and Shapiro, 2016).
A special type of cash transfers are grant or loan focused programs aimed at individ- uals who want to establish micro or small enterprises (MSEs) or who already run them.
While these programs vary in size and scope, they follow the rationale that the (poten- tial) entrepreneurs have good ideas and stamina, hence, might realize high returns to capital, but are financially constrained to borrow, save, or invest (Banerjee and Duflo, 2012; Dupas and Robinson, 2013b).3 Cash transfers or grants may alleviate these finan- cial frictions and help entrepreneurs grow their businesses and increase earned income (Blattman et al., 2018).4 Acknowledging the popularity and policy relevance of cash transfer programs, a substantial body of research discusses whether the mechanism of cash injections lifting capital constraints holds true, and if yes, under which conditions (see de Mel et al., 2008; McKenzie and Woodruff, 2008; de Mel et al., 2012; Fafchamps et al., 2014, among others). We build upon this existing work and examine whether a cash transfer of 300,000 Ugandan Shilling (UGX, about 100 PPP USD) to 96 micro and small entrepreneurs in Kampala, Uganda, affects their businesses’ profitability.
Adding to the well developed literature on cash transfers, we introduce a special twist in our randomized controlled trial: We additionally examine whether the impact of the cash transfer varies with themode of receiving it. Research in behavioral economics and psychology shows that people’s actions often hinge on the default option, i.e.
individuals do not make an active choice but silently agree to the default that is put in place for them (Thaler and Sunstein, 2008; Somville and Vandewalle, 2018). Following this logic, we test whether the mode (i.e. the default option) of receiving the unexpected cash transfer affects MSE owners differently due to their initial inertia. That is, half of our treatment group received the transfer directly in cash, while for the other half of
2 For a comprehensive and rigorous review of the literature consult Bastagli, Hagen-Zanker, Harman, Barca, Sturge, and Schmidt (Bastagli et al.).
3 Baird et al. (2018) coin the channel through which these transfers work the “self-employment liquidity effect”.
4 We omit discussing the literature on microcredit here although it is based on a similar mechanism.
the treatment group, we deposited the respective amount on their bank accounts. We assume that business owners who received the transfer in cash are more likely to spend the money right away as it is immediately available. Receiving the money on the bank account, however, might work as a soft commitment device for these entrepreneurs and help them to use it in a more organized manner, such as for business investments. All treated businesses are located in the urban centre of Kampala and bank branches or ATMs are nearby. Thus we expect transaction costs for the account treatment arm to be negligible. Any impact of the treatment for the latter group would consequently be due to the “default-option” effect of depositing it in the account and not due to high transaction costs. Following de Mel et al. (2008) and using detailed survey information on both entrepreneurs’ businesses and households, we study the impacts of the cash transfer on business profits and capital stock. Furthermore we investigate whether inventories, sales, and (business) savings change in response to the positive shock. As the randomized trial is embedded in a longer term project on micro-enterprise growth in sub-Saharan Africa, we are able to trace entrepreneurs not only in the short term (i.e.
6 months after they receive the cash grant), but up to four years after the intervention.
Beneficiaries report a significantly different usage of the transfer by treatment arm when asked about the subsequent use of the money in the short-term follow-up six months after the intervention: Entrepreneurs in the account treatment save signifi- cantly more of the transfer compared to the cash receivers. In contrast, those in the cash treatment arm report to invest significantly more money into their businesses.
These responses indicate that depositing the transfer on the bank account may possi- bly work as a soft-commitment device and that entrepreneurs in the cash treatment arm nevertheless use the money for business purposes despite money being fungible and the enumerators not framing what beneficiaries should do with the transfer.
Our intent-to-treat effects, however, do not mirror these self-reported survey results.
Although we find weak evidence for positive treatment impacts on upstream business variables such as inventories and sales for the account treatment group in the short term, they do not translate into sustained higher capital stocks or profits. In line with Fiala (2018), we do not detect any overall direct effect of the cash transfer on profits and capital stocks for any of the treatment arms, in the studied time period. Further, our heterogeneous treatment effects analysis reveals that men and women are not affected differently by the grant, neither in the cash nor in the account group. We find evidence for heterogeneous treatment effects only with respect to the level of education and baseline business savings of the entrepreneur. In both treatment arms, entrepreneurs with higher baseline business savings generate higher monthly sales. Interacting the
results in higher savings for entrepreneurs in both treatment groups. However as before, profits and capital stocks remain unaffected. Summarized, the way the money is handed out to recipients does not affect its subsequent use. Perhaps more importantly our (non-)results fail to provide evidence for the cash transfer to have an effect on micro and small enterprises. In line with a number of recent studies, our findings call into question the effectiveness of cash transfers as a means to spur growth in micro and small enterprises.
Literature
The paper relates to two main strands of literature: (i) We add to the extensive body of research on micro enterprise growth by presenting heterogeneous results with respect to the initial capabilities the entrepreneurs have to run their business successfully.
Furthermore we are able to show mid to long term results of cash transfers for micro and small businesses. (ii) Studying different modes of receiving the transfer allows us to examine whether the transfer on a bank account induces a default effect and hence influences financial decision making behavior.
(i) Cash transfer programs are typically seen as an effective tool to help firms over- come credit constraints and increase business profits in the short and medium term (e.g. de Mel et al., 2008, 2012). Among others, Blattman et al. (2016) vary the type of a cash transfer by additionally providing business training and supervision on top of a cash grant to “ultra-poor” women in Uganda and find large income gains after 18 months. In Sri Lanka and Ghana, micro and small business owners receive grants either in cash or in-kind (de Mel et al., 2012; Fafchamps et al., 2014). They result in positive effects on earnings for existing entrepreneurs in the short and medium term, whereby in Ghana, only female entrepreneurs benefit from the in-kind grant and there is no effect from providing cash. Other studies also fail to find overall significant ef- fects of cash transfers: Berge et al. (2015) for example state that cash grants are often consumed quickly instead of being used for investments by the entrepreneur. They find positive heterogeneous results of the intervention with the effect being more pro- nounced for male-led enterprises. Studying a grant, loan, and training intervention, Fiala (2018) cannot report any income effects from the grant treatment arm on neither female or male led businesses. He also fails to find an impact of any treatment for female businesses. A cash grant for young men engaged in petty crime in Monrovia, Liberia, also does not find lasting effects on business outcomes, which may be due to the dire conditions the men live in as the authors suggest (Blattman et al., 2017).
Newer studies explore whether cash transfer programs prevent the inter-generational
transmission of poverty. In a 10-year follow-up of a cash transfer in Ecuador, Araujo et al. (2017) report at most modest effects from the intervention on schooling outcomes of children. Blattman et al. (2014) and Haushofer and Shapiro (2016) present positive short-term results of two distinct cash transfer interventions on skilled self-employment and psychological well-being, respectively. Blattman et al. (2018) and Haushofer and Shapiro (2018), however, find that these effects dissipate over time. Blattman et al.
(2018) point out that the convergence effect of the control group over time is often neglected in short term studies. Cash transfers might just shift entrepreneurial invest- ments in time, but, as time goes by, non-treated entrepreneurs may also accumulate money and realize similar investments causing the initial effects on earned income to fade away. Our study adds to the longer term literature by accounting for changes up to four years after the intervention took place.
Another part of the literature on micro enterprise growth deals with the heterogene- ity of the informal sector beyond the effect of entrepreneurs’ gender. It distinguishes between a lower tier home to necessity entrepreneurs or survivalists and an upper tier consisting of high growth enterprises (Henrekson and Johansson, 2010). Grimm et al.
(2012) develop the concept of “constrained gazelles” entrepreneurs, i.e. business own- ers who exhibit high growth entrepreneurial potential (e.g. a high education, yet young age of the firm) but are limited by the lack of capital. This group tries to reach the upper tier but fails to do so. Evidently, exactly targeting these gazelles would improve the effectiveness of cash transfer programs. More recent studies succeed in identifying this special subgroup of entrepreneurs either via community panel decisions (Hussam et al., 2018) or through business plan competitions (Fafchamps and Woodruff, 2016;
Fafchamps and Quinn, 2017; McKenzie, 2017). Winners from these competitions are, among other things, more likely to be self-employed and to have more employees com- pared to those that rank second or third in the competition. Hence, cash transfers prove especially useful when targeted at the most promising group of (potential) en- trepreneurs. Our heterogeneous findings are in line with this evidence as we check for characteristics that determine successful entrepreneurship such as the level of education and the amount of savings.
(ii) The intervention studied here exploits variation in the mode of payment. We hypothesize that the deposit in the bank account works as a soft commitment device and hence enables us to study financial behavior. Research on savings behavior in particular finds that opposed to standard models of decision making, individuals tend to stick to the default option more often as is predicted (Choi et al., 2003). These
“default options” may be used to “nudge” people toward an outcome profitable for
research on 401(k) savings plans (Madrian and Shea, 2001) and on (default) organ decisions (Johnson and Goldstein, 2003). More similar to the context our work is based in, Fafchamps et al. (2014) detect a “flypaper effect” when in-kind transfers are made into small businesses in urban Ghana. While there is no effect on business outcomes for the group that received cash, the in-kind transfers remain in the business and result in increased profits for larger enterprises. A similar mechanism may play a role in our intervention. Business owners’ inertia may keep them from directly consuming the transfer deposited in the account and rather help them use it for productive purposes.
Lastly, our paper is related to work by Somville and Vandewalle (2018) and Brune et al. (2017) who study the effect of a cash transfer on savings behavior. These papers provide money either in cash or on a bank account to households in India and Malawi.
While Somville and Vandewalle (2018) confirm a direct and large impact of the repeated small transfers on household savings, Brune et al. (2017) find that a one-time-transfer only has limited effects on consumption and savings. Brune et al. (2017) explain these differing results by the fact that Somville and Vandewalle (2018) repeatedly deposit money in individual bank accounts and only find significantly different results after individuals have already received the transfers for several weeks. The work presented here examines a one-time cash transfer as in Brune et al. (2017) and differs from the two studies cited above mainly in terms of the target group. We specifically focus on business owners and do not only study the direct effects on the account savings balance but take into account business related outcomes such as profits, capital stock, sales, and business inventories.
The paper proceeds as follows: Section 2.2 introduces the study background and Section 2.3 the data used. It discusses the procedure of randomly allocating treatment and control groups, presents an attrition analysis and describes what respondents sub- jectively state to use the transfer for. Section 2.4 describes the estimation strategy along with main treatment results and heterogeneous treatment effects. Section 2.5 discusses limitations of our study and presents robustness tests before Section 2.6 concludes.
2.2 Background and Study Design
In this section we first briefly introduce the context the intervention was held in and then proceed to describe the experimental set-up in more detail.
Background
The study took place in urban Kampala which is the capital of Uganda. The country has seen considerable economic growth that slowed down from on average 7 percent annually during 1990-2010 to 4.5 percent p.a. in the five years prior to 2016, which also comprise the intervention period (World Bank, 2018b). In 2013, the time when the intervention set in, about 81 percent of the working-age population were self- employed, predominantly working in agriculture, followed by micro-enterprises. The latter cover about 90 percent of private sector production and employ over 2.5 million people (Fiala, 2018). While population growth and increased life expectancy will add more and more young people to the labor force in the future, they will face few wage employment opportunities and thus, self-employment is likely to expand. The majority of micro enterprises, however, face various constraints to business growth, of which lack of capital and missing skills are most important.
The intervention tackles capital constraints in particular and focuses on the dif- ference between transferring an unconditional grant directly in cash versus on a bank account. In 2017, 59 percent of Ugandans owned some form of formal bank account com- pared to 44 percent in 2014 and 20 percent in 2011 (Demirg¨u¸c-Kunt et al., 2015, 2018).
Account ownership has been boosted substantially by the spread of mobile money accounts which outweigh by far the sole ownership of a bank account at a financial in- stitution. Hence, provided that the account treatment really induces a soft-commitment effect, using the financial infrastructure via (mobile) cashless banking could promise an avenue for scalable interventions lifting capital constraints from business owners in the future.5
Experimental Design
The intervention was conducted by the German Institute of Global and Area Studies (GIGA, Hamburg) in collaboration with the Centre for Basic Research in Kampala.
5 Somville and Vandewalle (2018) motivate their research by efforts of the Indian government to spread cashless banking throughout the sub-continent. While they do not examine the policy’s consequences for micro enterprises, this paper shows possible implications for this sub-group.
and Small Firms in Developing Countries”. Listing of businesses for the final sample was carried out together with the first baseline survey in October 2012. In April 2013, a second baseline survey was conducted followed by the implementation of the treatment.
Trained enumerators interviewed the sample in a follow-up survey in October 2013 succeeded by further surveys in October 2014, 2015, 2016, and 2017. Entrepreneurs answered comprehensive questionnaires covering firm characteristics and performance, and provided information on personal and household characteristics including their financial situation and behavioral attitudes. We calculate a pooled treatment effect over all waves alongside a short-term effect including October 2013 and October 2014 and a rather long-term effect covering possible impacts from 2015 to 2017.
The overall baseline sample was drawn using a two-stage sampling procedure. A subset of this sample was eligible to participate in the intervention in either the con- trol or the treatment groups. It was drawn in the following way: First, 16 out of 220 geographical business zones in Kampala were randomly selected and all firms were listed therein.6 Second, 450 micro or small enterprises were randomly drawn strati- fied by industry branches.7 Finally, a random sub-sample of entrepreneurs was chosen to participate in the cash transfer intervention. In order to effectively capture micro and small enterprises, the random sample excluded enterprises with more than five employees and the ten percent enterprises with highest capital stocks (measured in October 2012). Table 2.1 shows the distribution of businesses across industries. Note that businesses in the “hair dressing and beauty” industry were oversampled in order to balance out male over-representation in branches such as “retail electric, phones, household appliances and related services” or “manufacturing”.
[Table 2.1 about here.]
To avoid a possible “double-intervention”, only banked entrepreneurs were eligible for the money transfer: The treatment effect would otherwise not be distinguishable from either opening a bank account per se or depositing the cash grant on the account had non-banked entrepreneurs been included in our sample draw. In fact, baseline data reveal that non-banked entrepreneurs exhibit significantly lower profits and capital
6 We applied cluster sampling: The zones were selected with a probability proportionate to the number of enterprises listed within them.
7 Industries were divided in “hair dressing and beauty”, “manufacture of printing/paper products and related services”, “manufacture of textile/wearing apparel (tailors) and related services”, “manu- facturing (remaining sectors)”, “retail and wholesale (remaining sectors”, “retail clothing, footwear, and leather”, “retail electric, phones, household appliances and related services”, and “other”.
stock, and lower levels of education. We therefore cannot draw conclusions on the effect of the transfer on the overall population but have to restrict our contribution to banked persons who, as already mentioned, represent more than half of the Ugandan population.
The experiment had two treatment arms. Originally, 50 businesses were randomly drawn to receive the cash transfer directly while 50 other enterprises were assigned to receive the grant on their bank account. 80 entrepreneurs were sampled for the banked control group. Several businesses had stopped working or shifted premises or could not be tracked for other reasons at the time when the intervention was to be realized (April 2013). These entrepreneurs were randomly replaced with other business owners that had answered the baseline survey from October 2012 already. Furthermore, three businesses differed in their intended and actual treatment status.8 Thus, in the main part of this paper, we estimateintent to treat estimates based on assignment to treatment in October 2012.
The final sample for our analysis is shown in Table 2.2.9 Differences in planned and actual number of respondents arise due to the following reasons: One entrepreneur was listed twice and randomly selected for both the cash and the account transfer.
We dropped these observations from our sample since receiving double the amount of money could distort our results. One business owner assigned for the account treatment was dropped from the sample because an employee instead of the actual owner had been interviewed. Furthermore, two entrepreneurs from the intended control and the cash groups, respectively, could not be interviewed at time of the intervention and did not get replaced. Last, baseline data for one entrepreneur in the account treatment group was deleted, because a wrong person had been interviewed. This explains why there is an additional respondent in the account treatment group in April 2013.
[Table 2.2 about here.]
The transfer amounted to 300,000 UGX which represents roughly monthly pre-intervention profits for the median firm in our sample. Receipt of the transfer was framed as remu- neration for participation in the survey. Business owners were free to use the money as
8 One entrepreneur was assigned to receive the account transfer, but turned out not to have a bank account. Another business owner that was assigned to receive an account treatment actually received the money in cash. The same happened for a person who was assigned for cash but who received the money via a bank account transfer. These business owners differed in their intended and actual treatment status.
9 Table A.1 in Appendix A shows the distribution of respondents who actually received the transfer or were actually included in the control group. Higher numbers in the control group arise due to more entrepreneurs having a bank account than previously stated at baseline in October 2012.
restricting the use of funds results in more honest reporting in subsequent waves. While the cash group received the money directly after they had responded to the April 2013 survey, the account treatment group received the money between one to fourteen days after the interview had taken place due to the enumerator team rechecking banking details and the banks’ need for time to transfer the money. The control group received the standard remuneration of 7,000 UGX that was regularly paid out after answering the questionnaire.
2.3 Data
This section explains how our main dependent variables are measured and deals with differential attrition over time. It also provides baseline summary statistics and survey results on what entrepreneurs state to use the cash transfer for.
2.3.1 Measurement of Main Variables
In line with de Mel et al. (2008), our main outcome variables are the firm’s monthly prof- its and capital stock. Basic treatment effects on these variables are calculated both in levels and logs to increase the robustness of the analysis. Furthermore, we test whether the transfer affected the amount of business inventories, sales, and total savings.
As de Mel et al. (2009b) recommend, profits were elicited directly from the respon- dent by asking a single survey question.10 According to them, this yields more precise estimates and performs better than calculating profits from detailed expenditure and revenue information. Field work experience confirms this view. Hence we are optimistic to use a realistic profit measure in our analysis. Similar to profits, total monthly sales are elicited directly from the entrepreneur.
Capital stock is derived from asking about business assets with respect to machinery, furniture, business tools, vehicles, land and business premises, and other remaining assets. Categorizing assets along these lines is supposed to help the entrepreneur recall assets as comprehensively as possible. Business owners did not only report the quantity but also the replacement value of assets at the time of the survey and whether they are rented or owned. These pieces of information are provided in each wave whereas the
10Enumerators ask the following: “What was the total profit the business made in the last four weeks after paying all expenses including wages of employees, raw materials, items for resale, electricity, water, fuel, rental etc.?” A second question elicits whether stated profits include the entrepreneurs personal income. If this is not the case, personal income is added to the profit measure.
replacement value of the former wave is repeated in the preceding survey as a built-in anchor. Values are corrected for if the respondents state that the past replacement value for a specific item is unrealistic. This is particularly done so for the October 2012 survey. While the replacement value accounts for asset depreciation, inflation is not considered. The capital stock measure we use in our analysis comprises the various asset groups except for the value of land and buildings because respondents find it hard to state a specific value and tend to overstate these values.
More specifically, inventories include the value of all items held as consumables, raw materials and finished goods. For the case of a carpenter for example, this means raw materials being wooden planks and finished goods being a bed, window frame, door etc.
As the value of all raw materials and finished goods is asked generally without precisely pinning down goods one by one, the values are likely prone to measurement error. How- ever, entrepreneurs preponderantly reported to use the cash transfer to buy inventories and equipment, so we include this measure in the regression analysis nevertheless.
Last, total savings amount to the sum of savings at home, savings on a bank account, on a mobile money account, at a savings club, and savings that are kept with neighbors or trusted friends. Business savings comprise only the money which is labeled for specific use in the business. All monetary measures are deflated to price levels in 2011 and stated in 1,000 UGX.
2.3.2 Sample Attrition
Short-term attrition between April and October 2013 was 3.4 percent while longer- term attrition up to the latest survey in 2017 amounts to 36.6 percent. Table 2.2 suggests that a higher share of business owners from the control group could not be interviewed in the follow-up waves contrary to the treated entrepreneurs. To test for differential attrition in our sample, we regress interview status in the follow-up round on treatment assignment for each wave separately and over all waves at once. We estimate the regression in the following form (see Dupas and Robinson, 2013b):
yi =β0+β1AccountT reatment+β2CashT reatment+i (2.1) The dependent variableyi equals 1 if the person could not be surveyed in the respective follow up wave,AccountT reatmentand CashT reatmentare dummy variables turning one if the respondent received the grant on a bank account, or in cash, respectively. The constantβ0 reflects the likelihood of the omitted treatment group, i.e. the respondents assigned to the control group, to not be followed-up. Table 2.3 depicts the results.
Entrepreneurs in both treatment arms were not more likely to be untracked in the follow-up waves than their control group counterpart. Even contrary, those who received the transfer in cash were more likely to be surveyed in the pooled regression over all waves (Table 2.3, column (6)). There is, however, differential attrition with respect to the control group as the coefficient for the constant is significantly positive throughout the waves (except for October 2016). Perhaps control entrepreneurs expected some kind of higher return for participation in the survey and refused to be interviewed upon noticing there was no reward besides the remuneration fee to be gained.
Business owners who left the sample exhibit different characteristics than those who stayed in the sample at baseline, i.e. during waves one and two. (for details see Table A.2 in Appendix A). They were generally better off than those who were interviewed in subsequent waves. Among other characteristics, these business owners were younger, more affluent, more likely to have a university degree and to speak English fluently.
Their firms were younger, they had higher start-up capital and savings while they generated more profits and higher sales. Acknowledging these differences, we apply Lee-bounds to our analysis in Section 2.5.
2.3.3 Summary Statistics and Balance Tests
We first present a short summary of important baseline characteristics for the whole sample before we turn to evaluate the randomization procedure. McKenzie (2012) shows that using multiple pre- and post treatment rounds in impact evaluations increases es- timate precision especially when outcome variables are likely to be noisy and relatively less auto-correlated such as - as he states - business profits, and household incomes and expenditures. Thus, we use data from the first two pre-treatment rounds as an average baseline and present summary statistics and results from balance tests regarding ran- domization into treatment and control groups in Tables 2.4, 2.5, and 2.6. Items that were only elicited in one of the pre-treatment waves are marked with an asterisk.
[Tables 2.4, 2.5, and 2.6 about here.]
At baseline, there are more male than female business owners in our sample (59 percent vs. 41 percent) who are mostly between 25 and 47 year old. 68 percent of entrepreneurs are married and the median household has 5 members. 95 percent of sampled business owners report to be literate. 10 percent did not finish primary school, 30 percent com- pleted primary school, 24 percent finished middle school, 18 percent completed their
A-levels and 15 percent completed university. The median business owner responds correctly to 7 out of 10 questions testing financial literacy and works about twelve hours a day for six days a week in the business. 84 percent of business owners are rather impatient, i.e. they would prefer to receive 20,000 UGX in one week’s time over receiving 30,000 UGX in five weeks’ time. Regarding firms at baseline, the median firm age is about 5 years, it has one employee, and median monthly profits range at 350,000 UGX while the median capital stock lies at 738,000 UGX. The median business sells items worth about 1,705,500 UGX a month and incurs costs of about 1,175,000 UGX which are split into costs for raw materials, finished goods, and other expenses such as business rent or electricity payments. Only 21 percent of all micro enterprises are formally registered with the Ugandan revenue authority.
There are significant differences with regard to some baseline characteristics which are most probably due to the small sample size.11 Specifically, entrepreneurs in the account treatment group are less likely to work in the trade sector, but more likely to work in the services or other business sectors (Table 2.4). Moreover, entrepreneurs in the account group have significantly more employees and work slightly less in their business than their control group counterpart. There is a higher share of business owners with completed O-levels in the account group. While their businesses had significantly less start-up capital at their command, their monthly value-added is higher than that of the control group.
Entrepreneurs in the cash treatment group also differ from control business owners in various ways (Table 2.5): On average, their households are significantly bigger and they are older than those in the control group. They are less likely to work in the central (and busy) division, but rather likely to operate in the Makindye Division a little more outside of the city center. They rather work in the services and other sectors as compared to the control group and - similar to the account treatment arm - there is a higher share of entrepreneurs with completed A-levels. Also, businesses in the cash treatment group produce more value added while having had less start-up capital at hand.
Lastly, cash and account treatment also exhibit significant differences with respect to the location of their work, the educational level, and entrepreneur’s age (Table 2.6). At least, differences between the two treatment arms are few and there are no significant different business characteristics. This is good news since we are especially
11One reason for the randomization failing partly might be that the randomization only happened between treatment and control and not between the different treatment arms and the control group.
Differences between the overall treatment and control group are less pronounced.
arms. However, acknowledging these differences, we control for baseline values of the respective variables in each of the subsequent regressions.
2.3.4 Self-Reported Evidence on Use of Cash Transfer
In the first follow-up wave in October 2013, six months after the intervention had taken place, business owners are asked to state what they used the money for. 10.4 percent (5 entrepreneurs) of those in the account group had not withdrawn any money yet. 30 out of 48 account entrepreneurs withdrew at least some money from the transfer 1-2 weeks after arrival. Three business owners in the account treatment saved the money for a while and withdrew only in August 2013 or later. In the cash treatment arm almost half of treated business owners (22 out of 46 persons) deposited between 40,000 UGX and 300,000 UGX at their bank accounts after receiving the transfer.
[Figures 2.1 and 2.2 about here.]
Figures 2.1 and 2.2 summarize the subjectively stated use of the monetary transfer by treatment group and by gender, respectively. Generally, entrepreneurs state to use the money predominantly to invest in business inventory and equipment for current or new businesses.
[Tables 2.7 and 2.8 about here.]
According to the subjective data, business owners in the cash group spend significantly more money on business inventory and equipment than treated account entrepreneurs (212,045 UGX vs. 167,917 UGX) (Table 2.7). Most entrepreneurs increase their ex- penditures on stock depending on the business sector (e.g. electrical appliances, hair products, clothes, fabric, timber, food). Some use the money as “top-up” and invest larger amounts in, e.g. new machinery for wood workshops. Expenditures also include costs for business rent, and electricity. Conversely, respondents in the account treatment group save significantly more money than those in the cash group (42,708 UGX vs.
9,090 UGX). There are no other significant differences between the treatment groups.
Asked for whether any of the entrepreneurs in the treatment groups introduce new or innovative products, two sided t-tests reveal no significant difference between the two treatment arms and between treatment and control group.
Splitting treated entrepreneurs by gender reveals that women invest higher amounts in business inventory and equipment for their current business as men while men invest
also in new businesses (Table 2.8). Contrary to our expectations, women do not state to use significantly more money for household expenditures than men. However, as money is fungible these descriptive data only provide some rough pattern on the possible use of the transfer.
2.4 Results
This section first explains the empirical strategy we employ before reporting the main results of the cash transfer. We then proceed to present results from a heterogeneity analysis with respect to the gender of the business owner, baseline educational status and baseline (business) savings.
2.4.1 Estimation Strategy
To test the hypotheses introduced in Section 2.1, we estimate the following intention to treat (ITT) model:
Yit =α+β1Accountit+β2Cashi,t +δt+θY¯i0+X
γiXi0+eit (2.2) where i refers to a specific business, t is time and Yi,t represents the outcome of interest. Accounti,t takes the value one from the first post-treatment round on if the business owner was assigned to receive the transfer on the bank account. Likewise, Cashi,t is a dummy variable which indicates whether the entrepreneur belongs to the cash treatment arm. We include ¯Yi,0 - the baseline outcome value of the dependent variable for firm i - in the regression as it increases statistical precision (McKenzie, 2012). Additionally, the matrix Xi,0 controls for those baseline characteristics that were different between treatment and control arms, i.e. the divisions and industries the businesses operate in, educational levels and age of the entrepreneur, initial wealth, the amount of start-up capital, the number of employees and firm age. In our analysis we make use of two available baselines and pool them to increase statistical power, because data on business outcomes are often found to be quite noisy (de Mel et al., 2009b).
Moreover, δt are wave fixed effects, and ei,t is the error term. All standard errors are clustered at the firm level and robust.
To increase statistical power we present results from a pooled regression over all waves as well as short term results only using data from the first two post-treatment waves (October 2013 and October 2014) and longer term results using data from the
include the long-term measure as recent research by e.g. Blattman et al. (2018) points out the need for longer term evaluation of cash transfers.
To further lower the impact of outliers on our results, we adapt the trimming proce- dure of de Mel et al. (2008) by trimming the top and bottom 5 percent of observations whose profits change most positively and most negatively between waves. We include both absolute as well as percentage changes in our measure.12
As already mentioned, we analyze heterogeneous treatment effects in a second step employing a similar estimation framework as above:
Yit=α+β1Accountit+β2Cashit+β3malei0+β4(Account∗malei,0) +β5(Cash∗malei0) +δt+θY¯i0+X
γiXi0+eit
(2.3)
The coefficientsβ4 andβ5display the interaction effect of being male and the respective treatment. The coefficientsβ1 andβ2then measure the average effect of having received the 300,000 UGX either on the bank account or cash on hands for treated women. The overall impact of receiving the money by treatment group for males is measured by β1+β4 and byβ2+β5, respectively. In further regressions, we exchange the interaction term male with baseline (business) savings and educational achievement.
2.4.2 Estimation of Basic Experimental Treatment Effects
We focus our main interest on business profitability variables in line with de Mel et al.
(2008) and test whether the cash grant had an impact on the firm’s capital stock and deflated monthly profits. Tables 2.9 and 2.10 present basic estimation results. Columns (1) to (3) in each of the respective tables show estimation results in levels, columns (4) to (6) depict results in logs.
[Tables 2.9 and 2.10 about here.]
Capital Stock and Profits Opposed to much of the literature on unconditional cash transfers, we do not find a short term effect of the grant on either treatment arm with respect to capital stock and monthly profits. Also, the various time dimensions do not reveal much: all coefficients are not statistically significantly different from zero, hence we cannot detect a possible short-term effect fading out in the longer term or vice versa, a longer term effect materializing when higher returns can be reaped from
12Running the regressions with an untrimmed sample reinforces the need for truncation as coefficients are unreasonably high and driven by outliers.
